[llvm] [TargetParser][AMDGPU] Fix getArchEntry(). (PR #69222)

[Ivan Kosarev via llvm-commits]

kosarev wrote:

> Is there any easy way to test it?

I don't see an easy way to cover that. It seems in LLVM we only call it from clang, but then the numeric id is something translated from a string id and therefore is always a known target. This was caught using amdgpu-dis, which relies on `getArchEntry()` via `getArchNameFromElfMach()`.

https://github.com/llvm/llvm-project/pull/69222